Domain-Regressive Continual Test-Time Adaptation

• This is a PyTorch/GPU Implementation of the paper Domain-Regressive Continual Test-Time Adaptation with Orthogonal Low-Rank Adapters. Our code is mainly based on the official PyTorch implementation of CoTTA.
• We have released the code about statistical characteristics collection on ViT based on the official PyTorch implementation of CFA
• We are committed to releasing the remaining code upon acceptance of our paper.

Dependencies

System

ubuntu 20.04
python 3.9.7
cuda 11.2

Packages

torch==1.10.0
torchvision==0.11.
timm==0.4.12

Environments

# It may take several minutes for conda to solve the environment
conda update conda
conda env create -f envs.yml
conda activate DRCTTA

Datasets

Dataset ImageNet-C can be downloaded from here.

DRCTTA

Domain regressive continual test-time adaptation on Transformers:

cd imagenet
bash run.sh

Domain regressive continual test-time adaptation on CNNs:

cd cifar
# cifar10
bash run_cifar10.sh
# cifar100
bash run_cifar100.sh

Collect statistical characteristics of features before LN layers of ViT:

cd collection
bash statistic.sh